google--adk-python/contributing/samples/core/artifacts

Artifacts Sample

Overview

This sample demonstrates how to use the Artifacts feature in ADK to handle different media types (image, audio, video), formats (text, HTML), and artifact versions. Artifacts allow agents to save large pieces of data or binary files outside the main conversation history to avoid cluttering the LLM context window. The agent can then load these artifacts when needed.

This sample showcases:

• Generating and saving a valid Image (BMP).
• Generating and saving a valid Audio file (WAV).
• Generating and saving a valid Video file (MP4) using OpenCV.
• Generating reports in both Text and HTML formats.
• Automatic handling of Artifact Versions.

Sample Inputs

• Generate a text report about AI agents

• Generate an HTML report about AI agents

• Generate a dummy image artifact

• Generate a dummy audio artifact

• Generate a dummy video artifact

• Load the latest version of the image artifact

Graph

graph TD
    START --> artifacts_agent

How To

1. Text and HTML Reports: The generate_report tool allows generating reports in either text or html format. It saves the content with the appropriate MIME type (text/plain or text/html) and file extension (.txt or .html).
2. Media Artifacts: The generate_media_artifact tool demonstrates how to save binary data (image, audio, video) as artifacts using types.Part.from_bytes and ctx.save_artifact.
   • Image: Generated manually as a valid 100x100 red BMP file using standard libraries.
   • Audio: Generated manually as a valid 1-second sine wave WAV file using standard libraries.
   • Video: Generated as a valid 3-second MP4 file with a moving square using OpenCV (codec avc1 for better compatibility).
3. Versioning: The generate_report tool creates a new version of the artifact every time it is called with the same format (text or html), as it uses fixed filenames (report.txt or report.html). The version number is returned to the user.
4. Loading Latest Version: The standard LoadArtifactsTool (available as load_artifacts to the model) can be used to load the latest version of any artifact, including media files.

Dependencies

To generate the video artifact, the sample requires opencv-python and numpy. You can install them with:

pip install opencv-python numpy

If these libraries are not installed, the tool will return a helpful error message explaining how to install them, while image and audio generation will still work.